AI ANALYSIS: JUSTICE FORGIVES A JUDGE WHO USED NONEXISTENT PRECEDENTS PRODUCED BY ARTIFICIAL INTELLIGENCE, WHILE LAWYERS ARE FINED, REPORTED TO BAR ASSOCIATIONS, AND SUBJECTED TO DISCIPLINARY PROCEEDINGS FOR THE SAME CONDUCT

 

JUSTICE FORGIVES A JUDGE WHO USED NONEXISTENT PRECEDENTS PRODUCED BY ARTIFICIAL INTELLIGENCE, WHILE LAWYERS ARE FINED, REPORTED TO BAR ASSOCIATIONS, AND SUBJECTED TO DISCIPLINARY PROCEEDINGS FOR THE SAME CONDUCT

Analysis by Márcia Almeida, with artificial intelligence assistance subject to human and documentary review

July 14, 2026

---

Watch the video of the Reporting Judge’s opinion here — audio in English and Spanish

TABLE OF CONTENTS

1. The case that exposed disciplinary inequality
2. What has been established by documentary evidence
3. What still remains concealed
4. The São Paulo Court of Justice’s justification: “We are learning”
5. The legal issue is not limited to the judge’s intent
6. Nonexistent precedents are not rhetorical ornaments
7. The underlying case cannot be deemed irrelevant without public scrutiny
8. The denial of oral argument and the possible denial of the right to be heard
9. The judge’s personal and nondelegable responsibility
10. CNJ Resolution No. 615/2025
11. The National Code of Judicial Ethics
12. LOMAN and judges’ official duties
13. The treatment imposed on lawyers in Brazil
14. Regional Labor Court of the 23rd Region: fine and referral to the Brazilian Bar Association
15. Paraná Court of Justice: bad-faith litigation and referral to the Brazilian Bar Association
16. Regional Labor Court of the 7th Region: false case law and violation of procedural ethics
17. Superior Court of Justice: referral to the Brazilian Bar Association because of possible AI hallucinations
18. The leading United States precedent: Mata v. Avianca
19. Other United States precedents
20. United Kingdom, Canada, and other countries
21. The objective comparison
22. The false distinction between “judicial error” and “official misconduct”
23. Intent is not a prerequisite for every form of disciplinary liability
24. Proportionality does not mean impunity
25. Judicial independence does not mean judicial irresponsibility
26. Violation of equality, impartial public administration, and legitimate expectations
27. Risks to objective impartiality
28. The institutional responsibility of the Office of the Chief Judicial Inspector and the Special Body
29. Necessary measures
30. Conclusion
31. References

1. THE CASE THAT EXPOSED DISCIPLINARY INEQUALITY

On July 14, 2026, the Brazilian legal news website Migalhas reported that the Special Body of the São Paulo State Court of Justice upheld the dismissal of a disciplinary complaint against a judge who used nonexistent judicial precedents generated by an artificial intelligence tool in a judicial decision.

The complaint was filed by the Brazilian Bar Association, São Paulo Section — OAB/SP.

According to the report, the Chief Judicial Inspector of the São Paulo judiciary, Appellate Judge Silvia Rocha, expressly acknowledged that:

“There was, in fact, an undesirable error by the judge in reproducing certain judicial precedents, certain decisions, generated by an artificial intelligence tool.”

The Reporting Judge also stated that the tool “hallucinated and created a judicial decision that did not exist.”

Despite this objective finding, she concluded that the incident did not justify the commencement of disciplinary administrative proceedings because, in her words:

“At this time, when we are learning how to use artificial intelligence, [the error] was not intentional and did not interfere with the analysis of the facts or the outcome of the adjudication.”

The Special Body followed this reasoning and denied the appeals filed against the dismissal.

The decision raises an unavoidable question:

Why is the argument that “we are learning” used to shield a judge from disciplinary liability, while lawyers who submit nonexistent precedents are fined, found liable for bad-faith litigation, reported to the Brazilian Bar Association, and subjected to disciplinary proceedings?

2. WHAT HAS BEEN ESTABLISHED BY DOCUMENTARY EVIDENCE

Consultation of the São Paulo Court of Justice’s official sources made it possible to identify the following elements:

2.1. Number of the disciplinary proceeding

The matter was processed under No.:

0000376-49.2026.2.00.0826

The numbering indicates an administrative proceeding registered within the Office of the Judicial Inspector/PJeCor environment.

2.2. Adjudication on July 1, 2026

At the administrative session of the Special Body held on July 1, 2026, Appellate Judge Silvia Rocha voted to deny the appeals.

The adjudication was adjourned at the request of Appellate Judge Campos Mello.

2.3. Final adjudication on July 8, 2026

At the session held on July 8, 2026, the Special Body unanimously denied the appeals.

Appellate Judge Campos Mello issued Concurring Opinion No. 87,263.

2.4. Lawyers who acted in the appeals

The official publications identify:

• Rosana Gibowski — OAB/SP Registration No. 136,957;
• Thais Pires de Camargo Rêgo Monteiro — OAB/SP Registration No. 205,657.

The participation of these professionals is consistent with the fact that the complaint involved the OAB/SP and a possible violation of the professional prerogatives of the legal profession.

The official source confirms proceeding No. 0000376-49.2026.2.00.0826, the opinion issued by Chief Judicial Inspector Silvia Rocha, the July 1 adjournment, the unanimous adjudication on July 8, and the participation of lawyers Rosana Gibowski and Thais Pires de Camargo Rêgo Monteiro.

TJSP · 2

The name of the judge and the original judicial proceeding were not disclosed in the public notices, in the Migalhas report, or in the indexed OAB/SP results. The full record is probably linked to PJeCor, and attributing the conduct to a named judge without access to the proceeding would be speculation. The Migalhas report also does not provide documentary proof of how the Office of the Judicial Inspector concluded that the false precedents did not affect the outcome.

Migalhas · 1

The addresses listed above were actually located and accessed during this research on July 14, 2026. Some judicial websites may subsequently change their routes, require a digital certificate or CAPTCHA, or block automated access.

2.5. Facts acknowledged by the Office of the Judicial Inspector itself

The Reporting Judge acknowledged:

1. that the judge used an artificial intelligence tool;
2. that the tool produced nonexistent judicial decisions;
3. that the false precedents were reproduced in the judicial decision;
4. that the judge had a duty to verify the cited authorities;
5. that such verification was not performed.

This is therefore not a mere journalistic suspicion. The use of nonexistent precedents was admitted during the administrative adjudication itself.

3. WHAT STILL REMAINS CONCEALED

Despite the institutional seriousness of the incident, the following information was not located on the public pages consulted:

• the judge’s name;
• the trial court or judicial unit in which the judge served;
• the number of the original judicial proceeding;
• the full nature of the underlying case;
• the decision containing the nonexistent citations;
• the complete list of fabricated precedents;
• the Office of the Judicial Inspector’s dismissal decision;
• the appeals submitted by the OAB/SP and by the lawyer;
• the full opinion issued by Appellate Judge Silvia Rocha;
• the full text of Concurring Opinion No. 87,263;
• the analytical demonstration that the false precedents did not influence the outcome.

This opacity prevents public scrutiny of a decisive conclusion: the assertion that the nonexistent precedents “did not interfere” with the adjudication.

Without access to the original decision and the false authorities, this assertion cannot be audited by society, the legal community, or the injured party.

4. THE SÃO PAULO COURT OF JUSTICE’S JUSTIFICATION: “WE ARE LEARNING”

The expression “we are learning” may justify training, institutional improvement, or the gradual adoption of new technologies.

By itself, however, it cannot eliminate preexisting legal duties.

Long before the widespread adoption of generative artificial intelligence, judges, lawyers, prosecutors, public defenders, and court personnel were already required to:

• verify the existence of the sources they cited;
• reproduce judicial decisions accurately;
• refrain from attributing nonexistent holdings to courts;
• refrain from inserting false facts into procedural acts;
• act in good faith, with loyalty and diligence;
• preserve the reliability of the administration of justice.

Artificial intelligence neither created nor revoked these duties.

The use of a new tool does not render acceptable conduct that, had it resulted from manual research, would be regarded as gross negligence.

5. THE LEGAL ISSUE IS NOT LIMITED TO THE JUDGE’S INTENT

The Office of the Judicial Inspector centered its reasoning on the absence of intent.

That reasoning is insufficient.

Disciplinary liability is not limited to deliberate fraud. It may also result from:

• gross negligence;
• recklessness;
• professional incompetence;
• breach of official duty;
• a lack of care incompatible with judicial office;
• conduct that undermines public confidence in the judiciary.

The fact that a judge did not intend to fabricate false case law does not eliminate the obligation to verify the content before signing and issuing the decision.

A judicial signature represents the personal assumption of authorship and responsibility for the judicial act.

The fact that the text may initially have been prepared by a law clerk, court employee, or artificial intelligence system does not sever the judge’s chain of responsibility.

6. NONEXISTENT PRECEDENTS ARE NOT RHETORICAL ORNAMENTS

A judicial decision is not merely a private opinion.

It is a coercive manifestation of state power and may:

• deprive a person of property;
• restrict liberty;
• dissolve family relationships;
• remove children from their parents;
• order evictions;
• deny medical treatment;
• uphold enforcement proceedings;
• terminate cases;
• foreclose appellate review;
• create res judicata.

When a judicial decision attributes to a court a holding that never existed, it presents as positive law what is merely an algorithmic fabrication.

The seriousness of the conduct does not disappear because the judge claims to have relied on other grounds.

Precedents are cited precisely in order to:

1. reinforce the authority of the conclusion;
2. demonstrate consistency with existing case law;
3. persuade the parties that the outcome is not arbitrary;
4. permit appellate review;
5. bind or guide future adjudications;
6. publicly legitimize the exercise of judicial power.

Had the precedents been entirely useless, they would not have been included in the decision.

7. THE UNDERLYING CASE CANNOT BE DEEMED IRRELEVANT WITHOUT PUBLIC SCRUTINY

The Reporting Judge stated that the nonexistent decisions did not interfere with the analysis of the facts or the outcome.

That conclusion requires a demonstration, not a mere assertion.

It would be necessary to compare:

• the genuinely existing independent grounds;
• the grounds supported by the false authorities;
• the substance of the dispute;
• the arguments presented by the party;
• the possible influence of the citations on the other judges;
• the subsequent use of the decision in other proceedings;
• the possibility that the party was induced to abandon or alter its appellate strategy.

Without publication of the underlying decision and the disciplinary opinion, the premise adopted by the court cannot be verified.

Moreover, even if the same outcome could have been reached on other grounds, false legal information would still have been inserted into an official state act.

8. THE DENIAL OF ORAL ARGUMENT AND THE POSSIBLE DENIAL OF THE RIGHT TO BE HEARD

The disciplinary complaint did not concern only the nonexistent case law.

According to the report, it also challenged the denial of a request for synchronous oral argument.

This fact increases the seriousness of the context.

The party simultaneously alleged:

1. that the judge used precedents that did not exist;
2. that her lawyer was not permitted to present synchronous oral argument;
3. that there was a possible denial of the right to be heard and to present a defense.

The disciplinary analysis fragmented these issues and classified them as purely jurisdictional matters.

However, the combination of reasoning based on false authorities and restrictions on the exercise of the right to present a defense may reveal a structural due process problem.

Oral argument is not a favor granted to counsel. It is an instrument of participation, adversarial proceedings, and effective influence on the formation of the judicial decision.

The OAB/SP itself has published studies arguing that, once requested, synchronous oral argument must be guaranteed, particularly when a virtual or asynchronous adjudication eliminates real-time interaction between counsel and the judges.

9. THE JUDGE’S PERSONAL AND NONDELEGABLE RESPONSIBILITY

Artificial intelligence may assist in organizing information, linguistic review, preliminary research, and preparation of draft decisions.

It cannot assume the constitutional function of adjudication.

A judge who signs a decision institutionally declares that he or she:

• examined the case file;
• knows the grounds employed;
• verified the authorities;
• agrees with the content;
• assumes legal and ethical responsibility for the ruling.

The response that “the artificial intelligence hallucinated” describes the mechanism of the error, but it does not eliminate human responsibility.

AI does not hold public office, has not passed a competitive public examination, possesses no decisional independence, takes no constitutional oath, and cannot be subjected to disciplinary liability.

Responsibility must rest with the person who decided to use its output without verification.

10. CNJ RESOLUTION NO. 615/2025

CNJ Resolution No. 615 of March 11, 2025, establishes rules governing the development, governance, auditing, monitoring, and responsible use of artificial intelligence within the Brazilian judiciary.

The Resolution requires compliance with the principles of:

• transparency;
• explainability;
• contestability;
• auditability;
• reliability;
• security;
• human oversight;
• protection of fundamental rights.

The Resolution also requires secure, traceable, and auditable sources.

A nonexistent judicial precedent is not:

• secure;
• traceable;
• auditable;
• reliable;
• explainable;
• compatible with the duty of transparency.

The matter should therefore not have been examined merely as an individual failure in learning to use technology, but as a possible violation of the national rules governing artificial intelligence in the judiciary.

11. THE NATIONAL CODE OF JUDICIAL ETHICS

The National Code of Judicial Ethics imposes upon judges duties of:

• independence;
• impartiality;
• knowledge and professional competence;
• courtesy;
• transparency;
• prudence;
• diligence;
• professional and personal integrity;
• dignity, honor, and decorum.

Prudence requires a judge to rationally assess the consequences of judicial decisions.

Diligence requires careful and timely performance of judicial duties.

Integrity requires consistency between the extensive authority entrusted to judicial office and the reliability of the information contained in judicial acts.

The insertion of AI-fabricated case law may simultaneously violate all three duties.

12. LOMAN AND JUDGES’ OFFICIAL DUTIES

Article 35 of the Organic Law of the National Judiciary — Complementary Law No. 35/1979 — establishes, among other duties, the obligation to:

• comply with and enforce statutory provisions and official acts;
• refrain from unjustifiably exceeding procedural deadlines;
• take the measures necessary to ensure that procedural acts are performed regularly;
• treat the parties, lawyers, and other participants with courtesy;
• maintain irreproachable conduct in public and private life.

Procedural regularity includes the issuance of legally verifiable decisions.

A state ruling that presents as existing a judicial decision that was never issued is not procedurally regular.

13. THE TREATMENT IMPOSED ON LAWYERS IN BRAZIL

While the São Paulo Court of Justice concluded that the judge was “learning,” Brazilian courts have imposed the following consequences on lawyers in similar circumstances:

• fines for bad-faith litigation;
• liability for procedural expenses;
• refusal to entertain appeals;
• dismissal of actions;
• referral to the Brazilian Bar Association;
• orders for disciplinary investigation;
• findings of violation of procedural good faith;
• accusations of attempting to mislead the judiciary.

It is legitimate to hold professionals accountable for submitting false precedents.

The problem lies in the absence of an equivalent standard for those exercising judicial authority.

14. REGIONAL LABOR COURT OF THE 23RD REGION: FINE AND REFERRAL TO THE BRAZILIAN BAR ASSOCIATION

In July 2026, the Regional Labor Court of the 23rd Region, in the State of Mato Grosso, reported a case in which five nonexistent precedents attributed to the Superior Labor Court and other courts were identified.

The court held that submitting false references violated the duties of loyalty, caution, and procedural good faith.

The court ordered:

• the imposition of a fine;
• referral to the Brazilian Bar Association for investigation of the professional conduct.

The official report issued by the Regional Labor Court of Mato Grosso did not adopt the rationale that the lawyers were “learning” to use artificial intelligence.

The failure to verify the authorities was treated as professional responsibility.

Summary of the holding

ARTIFICIAL INTELLIGENCE. NONEXISTENT PRECEDENTS. FAILURE OF PROFESSIONAL VERIFICATION. BAD-FAITH LITIGATION. The use of artificial intelligence tools without human supervision does not relieve counsel of responsibility for the contents of a filing. The submission of nonexistent precedents violates the duties of good faith, loyalty, and caution, authorizing the imposition of a fine and referral to the Brazilian Bar Association.

15. PARANÁ COURT OF JUSTICE: BAD-FAITH LITIGATION AND REFERRAL TO THE BRAZILIAN BAR ASSOCIATION

The Paraná State Court of Justice examined an interlocutory appeal in which nonexistent precedents, apparently produced by AI, were cited.

The appellate decision held that:

“By using an artificial intelligence tool without the necessary review and verification, the appellants violated their duty of care.”

It also concluded that the submission of nonexistent case law constituted conduct capable of misleading the judiciary.

The following measures were upheld or ordered:

• a fine for bad-faith litigation;
• an official referral to the Paraná Section of the Brazilian Bar Association;
• investigation of the professional conduct.

Essential headnote — Paraná Court of Justice

INTERLOCUTORY APPEAL. USE OF ARTIFICIAL INTELLIGENCE. CITATION OF NONEXISTENT PRECEDENTS. VIOLATION OF THE DUTY OF CARE. BAD-FAITH LITIGATION. REFERRAL TO THE BRAZILIAN BAR ASSOCIATION. The use of an artificial intelligence tool without human review and the submission of nonexistent judicial decisions violate procedural good faith and justify the imposition of a fine and referral to the Brazilian Bar Association.

In another adjudication, the Paraná Court of Justice imposed a fine equivalent to one minimum monthly wage, although it acknowledged that the false precedents did not affect the principal issue raised in the appeal.

This fact is especially important.

In the Paraná case, the fact that the false case law was not outcome-determinative did not eliminate the sanction.

In the case of the São Paulo judge, the same argument was used to prevent the commencement of any disciplinary proceeding.

16. REGIONAL LABOR COURT OF THE 7TH REGION: FALSE CASE LAW AND VIOLATION OF PROCEDURAL ETHICS

In June 2025, the Regional Labor Court of the 7th Region reported a finding of bad-faith litigation against a lawyer who submitted nonexistent or altered case law.

The court emphasized that the use of nonexistent or deliberately modified precedents violates fundamental principles of procedural ethics.

It also ordered referral of the matter to the Ceará Section of the Brazilian Bar Association.

Summary of the holding

NONEXISTENT OR ALTERED CASE LAW. ATTEMPT TO MISLEAD THE COURT. BAD-FAITH LITIGATION. The submission of false precedents undermines confidence in the judicial process, violates procedural loyalty, and authorizes sanctions, without prejudice to disciplinary investigation by the Brazilian Bar Association.

17. SUPERIOR COURT OF JUSTICE: REFERRAL TO THE BRAZILIAN BAR ASSOCIATION BECAUSE OF POSSIBLE AI HALLUCINATIONS

In May 2026, the Brazilian Superior Court of Justice reported a decision issued by Justice Rogerio Schietti Cruz in a habeas corpus proceeding.

The Reporting Justice identified serious defects in the application, including:

• incorrect citations;
• nonexistent judicial decisions;
• material apparently produced by artificial intelligence without verification.

In addition to denying interim relief, he ordered that the matter be referred to the Brazilian Bar Association.

Once again, the absence of deliberate intent did not prevent the adoption of disciplinary measures.

18. THE LEADING UNITED STATES PRECEDENT: MATA v. AVIANCA

The best-known international precedent is Mata v. Avianca, Inc., decided in 2023 by the United States District Court for the Southern District of New York.

The lawyers submitted a brief containing nonexistent decisions produced by ChatGPT.

When the court questioned the references, they provided fabricated texts purporting to be complete judicial opinions.

Judge P. Kevin Castel concluded that the lawyers had abandoned their gatekeeping function and submitted to the court:

• nonexistent judicial opinions;
• false citations;
• fabricated quotations;
• misleading statements.

The court imposed financial sanctions of US$5,000, together with other measures, including notification of the judges falsely identified as authors of the fabricated decisions and notification of the client.

The court emphasized that there is no absolute prohibition against using artificial intelligence, but that lawyers serve a gatekeeping role and must ensure the accuracy of everything they submit.

Central passage of the decision

There is nothing inherently improper about using a reliable artificial intelligence tool for assistance. Existing rules, however, impose a gatekeeping role on attorneys to ensure the accuracy of their filings.

The court also listed the harms caused by fabricated decisions:

• wasting the opposing party’s time;
• wasting the court’s time;
• harming the client;
• damaging the reputations of judges falsely identified as authors;
• damaging the reputations of parties falsely associated with fictional facts;
• eroding confidence in the judicial process.

These harms apply equally when the person responsible for the act is a judge.

In fact, the harm may be even greater because the false authority appears within an official state decision carrying a presumption of legitimacy.

19. OTHER UNITED STATES PRECEDENTS

Following Mata v. Avianca, United States federal and state courts began imposing or considering:

• monetary sanctions;
• costs and attorneys’ fees;
• rejection or striking of filings;
• orders to show cause;
• disciplinary referrals;
• formal admonishments;
• certification requirements confirming that cited authorities were verified.

In Garner v. Kadince, the Utah Court of Appeals reaffirmed that artificial intelligence does not alter the professional obligation to verify legal authorities.

In Cassata v. Michael Macrina Architect, P.C., a New York court observed that a substantial body of decisions already recognizes the following conduct as sanctionable:

• citing nonexistent cases;
• using nonexistent quotations;
• attributing false propositions of law to real cases.

In federal decisions issued in 2026, courts warned that the problem has evolved: AI does not always fabricate an entire case; it frequently attributes to a real case a quotation, legal proposition, or outcome that the case never contained.

This form of falsity is even more difficult to detect.

20. UNITED KINGDOM, CANADA, AND OTHER COUNTRIES

20.1. United Kingdom

English courts have warned that lawyers have a professional duty to verify every authority they submit.

In cases involving false AI-generated case law, courts have considered:

• contempt of court;
• the lawyer’s personal liability;
• referral to professional regulatory bodies;
• adverse costs orders;
• possible interference with the administration of justice.

The fundamental position is that unfamiliarity with the technology is not a sufficient excuse.

20.2. Canada

In Canada, cases involving negligent use of AI have already resulted in:

• adverse costs awards;
• special costs imposed against lawyers;
• proceedings before professional law societies;
• orders requiring reimbursement of expenses caused by false research.

In the disciplinary proceeding involving lawyer Javad Mazaheri, the negligent use of artificial intelligence contributed to increasing the complexity of the proceeding. The panel imposed costs of approximately CAD 31,150 in favor of the professional regulatory body.

20.3. International trend

The comparative international trend is clear:

1. the use of AI is not prohibited;
2. responsibility remains human;
3. failure to verify is subject to censure;
4. intentional misconduct may aggravate the violation but is not a prerequisite for every consequence;
5. prompt correction and transparency may mitigate the sanction;
6. an attempt to conceal the error aggravates responsibility;
7. false authorities undermine the integrity of justice.

21. THE OBJECTIVE COMPARISON

Element	Sanctioned lawyers	São Paulo judge
Use of AI	Acknowledged or inferred	Acknowledged
Nonexistent precedents	Yes	Yes
Failure to verify	Treated as a professional violation	Expressly acknowledged
Requirement of intent	Frequently unnecessary for a fine or referral to the Bar	Absence of intent used to dismiss the matter
Effect on the merits	Not always required	Alleged lack of effect used to avoid investigation
Consequence	Fine, bad-faith finding, costs, Bar referral, disciplinary proceedings	No disciplinary proceedings
“Learning” justification	Generally rejected	Accepted
Transparency	Proceedings and headnotes frequently public	Judge’s name and original case concealed

The difference is not merely quantitative.

It is a difference of principle.

For a lawyer, failure to review constitutes sanctionable negligence.

For a judge, the same failure to review was treated as a tolerable stage in the learning process.

22. THE FALSE DISTINCTION BETWEEN “JUDICIAL ERROR” AND “OFFICIAL MISCONDUCT”

It is correct to state that the Office of the Judicial Inspector cannot operate as an appellate tribunal to correct every mistaken interpretation of law.

This case, however, does not involve only an interpretive disagreement.

The question was not whether the judge correctly interpreted a real precedent.

The issue was the use of precedents that did not exist.

There is an essential difference between:

• misinterpreting a real decision; and
• representing an imaginary decision as real.

The first may constitute judicial error.

The second may constitute a lack of diligence, violation of the duty of truthfulness, noncompliance with AI-governance rules, and impairment of public confidence.

Automatically classifying the episode as a jurisdictional matter creates a zone of disciplinary immunity for any falsehood inserted into the reasoning of a judicial decision.

23. INTENT IS NOT A PREREQUISITE FOR EVERY FORM OF DISCIPLINARY LIABILITY

The absence of fraudulent intent is relevant to the determination of the appropriate sanction.

It does not necessarily eliminate the violation.

A rational disciplinary system distinguishes among:

• an isolated clerical error that is immediately corrected;
• ordinary negligence;
• gross negligence;
• repeated conduct;
• deliberate blindness;
• knowing falsity;
• an attempt to conceal the conduct.

The commencement of disciplinary proceedings is not equivalent to a finding of guilt.

Its purpose is precisely to determine:

1. which tool was used;
2. who prepared the draft;
3. which prompts or instructions were entered;
4. how many precedents were false;
5. whether any verification was performed;
6. whether the decision was corrected;
7. whether the parties were formally notified;
8. whether other cases were affected;
9. whether the conduct was repeated;
10. whether confidential information was transmitted to an external system;
11. whether the tool was authorized by the court;
12. whether CNJ Resolution No. 615/2025 was violated.

Dismissing the matter before conducting that investigation means replacing fact-finding with a presumption of harmlessness.

24. PROPORTIONALITY DOES NOT MEAN IMPUNITY

The principles of reasonableness and proportionality do not require every error to result in the most severe sanction.

Nor do they authorize treating every judicial failure as irrelevant.

Between removal of the judge from office and outright dismissal of the complaint, several possible responses exist:

• admonishment;
• censure;
• formal recommendation;
• mandatory training;
• auditing of decisions produced with AI assistance;
• public retraction or correction;
• notice to the affected parties;
• review of other similar judicial acts;
• temporary prohibition on the use of external tools;
• a requirement that the use of AI be transparently documented;
• binding institutional guidance.

Proportionality requires a graduated response.

It does not require the absence of any response.

25. JUDICIAL INDEPENDENCE DOES NOT MEAN JUDICIAL IRRESPONSIBILITY

Judicial independence protects judges against improper pressure.

It does not protect:

• fabrication of sources;
• technological negligence;
• failure to verify;
• irresponsible use of systems;
• concealment of the decision-making method;
• violation of procedural rights.

Judicial independence exists for the benefit of litigants and the rule of law, not as a personal privilege of the judge.

The greater the governmental power, the greater the responsibility governing its exercise.

26. VIOLATION OF EQUALITY, IMPARTIAL PUBLIC ADMINISTRATION, AND LEGITIMATE EXPECTATIONS

Article 5 of the Brazilian Constitution guarantees equality before the law.

Article 37 subjects public administration to the principles of legality, impartiality, morality, publicity, and efficiency.

Although the legal regimes governing judges and lawyers are not identical, functionally equivalent situations must be assessed according to coherent standards.

The submission of a nonexistent legal authority creates the same epistemic risk regardless of who submits it:

• a lawyer;
• a prosecutor;
• a public defender;
• a court employee;
• a trial judge;
• an appellate judge;
• a justice of a higher court.

When the system severely punishes a person who attempts to persuade a judge with false case law, but tolerates the judge’s reliance on the same falsehood in the judicial decision itself, it conveys the message that procedural truth depends on the official position of the person who makes the error.

This destroys legitimate confidence in institutional neutrality.

27. RISKS TO OBJECTIVE IMPARTIALITY

Impartiality is not limited to the absence of a personal interest.

It also requires the public appearance of neutrality and equal treatment.

A reasonable observer may ask:

• why the judge received institutional understanding;
• why lawyers receive fines and disciplinary referrals;
• why the judge’s name remained protected;
• why the false decisions were not publicly identified;
• why no transparent audit was conducted;
• why the same Bar Association that files complaints against lawyers was unsuccessful when it requested an investigation of a judge.

The appearance of double standards is itself harmful to the legitimacy of the judiciary.

28. THE INSTITUTIONAL RESPONSIBILITY OF THE OFFICE OF THE JUDICIAL INSPECTOR AND THE SPECIAL BODY

The issue extends beyond the individual conduct of the judge.

By acknowledging the use of nonexistent case law and nevertheless dismissing the complaint on the ground that the judiciary is learning to use the technology, the court established an administrative precedent.

That precedent may be invoked in the future by other judges.

For the sake of consistency, lawyers could also argue:

• that they were learning;
• that the AI hallucinated;
• that they lacked intent;
• that the legal position could have been supported on other grounds;
• that the false reference did not change the outcome.

If these arguments are rejected when raised by lawyers, the court must objectively explain why they were sufficient when raised on behalf of a judge.

Institutional credibility requires an answer.

29. NECESSARY MEASURES

In light of the public interest involved, the following measures would be justified:

29.1. Publication of the opinions

The São Paulo Court of Justice should make available, subject to legally required safeguards:

• the opinion issued by Appellate Judge Silvia Rocha;
• Concurring Opinion No. 87,263;
• the full reasoning supporting dismissal of the matter.

29.2. Identification of the underlying judicial proceeding

The number of the original judicial proceeding should be disclosed unless a concrete and legally justified confidentiality order applies.

Even in a confidential case, it would be possible to anonymize the parties without concealing:

• the procedural classification;
• the judicial unit;
• the content of the false authorities;
• the nature of the error;
• the corrective measures adopted.

29.3. Publication of the nonexistent precedents

The court must disclose:

• which cases were fabricated;
• which docket numbers were false;
• which courts were improperly identified;
• which legal propositions were attributed to the imaginary decisions.

29.4. Audit

Other decisions produced through the same method should be audited, while preserving the presumption of regularity but determining whether the conduct was repeated.

29.5. Notice to the parties

The parties to the original case should receive formal notice and an opportunity to request:

• correction;
• retraction;
• a new adjudication;
• reopening of a procedural deadline;
• clarification;
• preservation of technological records.

29.6. Equal disciplinary standards

The National Council of Justice and the courts should adopt a national protocol establishing equivalent standards for judges, lawyers, prosecutors, public defenders, and court personnel.

29.7. Documentation of AI use

Every judicial decision produced with substantial assistance from generative AI should contain an auditable internal record identifying:

• the tool used;
• the date;
• the version;
• the purpose;
• the person responsible for the review;
• confirmation that the sources were verified.

30. CONCLUSION

The problem is not the use of artificial intelligence.

The problem is the substitution of technological excuses for human responsibility.

When a lawyer submits nonexistent case law, courts correctly state that the tool did not sign the filing and that the professional is responsible for verifying its content.

The same principle must apply to judges.

AI did not sign the São Paulo judge’s decision.

AI does not exercise judicial authority.

AI possesses no decisional independence.

AI cannot impose obligations, restrict rights, or create res judicata.

The person who signed the decision was a state official invested with enormous constitutional authority.

For society to trust judicial decisions, it must have the minimum assurance that the cases cited in those decisions actually exist.

Dismissal without a transparent investigation, based on the assertion that “we are learning,” stands in direct contrast to the fines, findings of bad faith, and disciplinary referrals imposed on lawyers for the same errors.

Equality before the law cannot end at the door of the Office of the Judicial Inspector.

Judicial integrity requires a simple rule:

The same artificial intelligence cannot be treated as evidence of negligence when used by a lawyer and as a mere learning tool when used by a judge.

Without consistency, transparency, and accountability, technological innovation does not modernize justice.

It merely automates and expands its inequalities.

31. REFERENCES

31.1. São Paulo Court of Justice case

MIGALHAS. Judge cites nonexistent case law and the São Paulo Court of Justice imposes no punishment: “We are learning.” São Paulo, July 14, 2026. Available at:
https://www.migalhas.com.br/quentes/460324/juiz-cita-jurisprudencia-inexistente-e-corregedora-vota-contra-punicao
Accessed on: July 14, 2026.

SÃO PAULO. Tribunal de Justiça do Estado de São Paulo [São Paulo State Court of Justice]. Agenda and Composition of the Special Body — July 1, 2026. Administrative proceeding No. 0000376-49.2026.2.00.0826. São Paulo, June 19, 2026. Available at:
https://www.tjsp.jus.br/OrgaoEspecial/Comunicados/Comunicado?codigoComunicado=70024&pagina=1
Accessed on: July 14, 2026.

SÃO PAULO. Tribunal de Justiça do Estado de São Paulo [São Paulo State Court of Justice]. Results of the Special Body — July 1, 2026. Administrative proceeding No. 0000376-49.2026.2.00.0826. São Paulo, July 3, 2026. Available at:
https://www.tjsp.jus.br/OrgaoEspecial/Comunicados/Comunicado?codigoComunicado=71364&pagina=1
Accessed on: July 14, 2026.

SÃO PAULO. Tribunal de Justiça do Estado de São Paulo [São Paulo State Court of Justice]. Agenda and Composition of the Special Body — July 8, 2026. Administrative proceeding No. 0000376-49.2026.2.00.0826. São Paulo, June 30, 2026. Available at:
https://www.tjsp.jus.br/OrgaoEspecial/Comunicados/Comunicado?codigoComunicado=70930&pagina=1
Accessed on: July 14, 2026.

SÃO PAULO. Tribunal de Justiça do Estado de São Paulo [São Paulo State Court of Justice]. Results of the Special Body — July 8, 2026. Administrative proceeding No. 0000376-49.2026.2.00.0826. São Paulo, July 13, 2026. Available at:
https://www.tjsp.jus.br/OrgaoEspecial/Comunicados/Comunicado?codigoComunicado=71754&pagina=1
Accessed on: July 14, 2026.

SÃO PAULO. Tribunal de Justiça do Estado de São Paulo [São Paulo State Court of Justice]. Special Body — Session of July 1, 2026. YouTube. Available at:
https://www.youtube.com/@tjspoficial
Accessed on: July 14, 2026.

SÃO PAULO. Tribunal de Justiça do Estado de São Paulo [São Paulo State Court of Justice]. Special Body — Session of July 8, 2026. YouTube. Available at:
https://www.youtube.com/watch?v=LOKQsQIxolc
Accessed on: July 14, 2026.

ORDEM DOS ADVOGADOS DO BRASIL — SEÇÃO DE SÃO PAULO [BRAZILIAN BAR ASSOCIATION — SÃO PAULO SECTION]. Studies on Virtual Hearings and Adjudications. São Paulo, 2026. Available at:
https://www.oabsp.org.br/upload/4232881259.pdf
Accessed on: July 14, 2026.

31.2. Brazilian rules governing artificial intelligence and judicial responsibility

BRAZIL. Conselho Nacional de Justiça [National Council of Justice]. Resolution No. 615 of March 11, 2025. Establishes rules governing the development, governance, auditing, monitoring, and responsible use of artificial intelligence solutions in the judiciary. Brasília, Federal District: CNJ, 2025. Available at:
https://atos.cnj.jus.br/atos/detalhar/6001
Accessed on: July 14, 2026.

BRAZIL. Conselho Nacional de Justiça [National Council of Justice]. National Code of Judicial Ethics. Brasília, Federal District: CNJ. Available at:
https://www.cnj.jus.br/codigo-de-etica-da-magistratura/
Accessed on: July 14, 2026.

BRAZIL. Complementary Law No. 35 of March 14, 1979. Provides for the Organic Law of the National Judiciary. Brasília, Federal District: Office of the President of the Republic. Available at:
https://www.planalto.gov.br/ccivil_03/leis/lcp/lcp35.htm
Accessed on: July 14, 2026.

BRAZIL. Conselho Nacional de Justiça [National Council of Justice]. The Use of Generative Artificial Intelligence in the Brazilian Judiciary. Brasília, Federal District: CNJ, 2024. Available at:
https://www.cnj.jus.br/wp-content/uploads/2024/09/cnj-relatorio-de-pesquisa-iag-pj.pdf
Accessed on: July 14, 2026.

31.3. Brazilian decisions and cases involving lawyers

BRAZIL. Tribunal Regional do Trabalho da 23ª Região [Regional Labor Court of the 23rd Region]. Regional Labor Court of Mato Grosso Fines Lawyer for Using False Precedents and Requests Investigation by the Brazilian Bar Association. Cuiabá, July 2026. Available at:
https://portal.trt23.jus.br/portal/noticias/trtmt-multa-advogado-por-uso-de-precedentes-falsos-e-solicita-apuracao-na-oab
Accessed on: July 14, 2026.

PARANÁ. Tribunal de Justiça do Estado do Paraná [Paraná State Court of Justice]. Appellate decision in Case No. 0032636 — nonexistent precedents produced by artificial intelligence, bad-faith litigation, and referral to the Paraná Section of the Brazilian Bar Association. Curitiba, June 25, 2025. Available at:
https://portal.tjpr.jus.br/jurisprudencia/j/4100000032876741/Ac%C3%B3rd%C3%A3o-0032636-
Accessed on: July 14, 2026.

PARANÁ. Tribunal de Justiça do Estado do Paraná [Paraná State Court of Justice]. Interlocutory Appeal No. 0147209-78.2025.8.16.0000 — improper use of artificial intelligence, nonexistent precedents, fine, and referral to the Brazilian Bar Association. Curitiba, 2026. Available at:
https://portal.tjpr.jus.br/jurisprudencia/j/4100000036373321/Ac%C3%B3rd%C3%A3o-0147209-78.2025.8.16.0000
Accessed on: July 14, 2026.

BRAZIL. Tribunal Regional do Trabalho da 7ª Região [Regional Labor Court of the 7th Region]. Lawyer Who Fabricated Case Law Is Found Liable for Bad-Faith Litigation and the Case Is Referred to the Ceará Bar Association. Fortaleza, June 12, 2025. Available at:
https://www.trt7.jus.br/index.php/noticias/todas-as-noticias/15649-advogado-que-falsificou-jurisprudencia-e-condenado-por-litigancia-de-ma-fe-e-caso-vai-a-oab-ce
Accessed on: July 14, 2026.

BRAZIL. Superior Tribunal de Justiça [Superior Court of Justice]. Decision Identifying Possible AI Defects in a Habeas Corpus Application Is Featured in STJ News. Brasília, Federal District, May 26, 2026. Available at:
https://www.stj.jus.br/sites/portalp/Paginas/Comunicacao/Noticias/2026/26052026-Decisao-que-apontou-possiveis-falhas-de-IA-em-pedido-de-habeas-corpus-e-destaque-no-STJ-Noticias.aspx
Accessed on: July 14, 2026.

BRAZIL. Justiça do Trabalho [Labor Judiciary]. Case No. 0000433-19.2025.5.06.0201 — artificial intelligence and the use of nonexistent precedents. Falcão System. Available at:
https://jurisprudencia.jt.jus.br/jurisprudencia-nacional/pesquisa/numero/0000433-19.2025.5.06.0201?abaSelecionada=acordaos
Accessed on: July 14, 2026.

31.4. United States

UNITED STATES. United States District Court for the Southern District of New York. Mata v. Avianca, Inc., 678 F. Supp. 3d 443, No. 1:22-cv-01461-PKC, Document 54. New York, June 22, 2023. Available at:
https://law.justia.com/cases/federal/district-courts/new-york/nysdce/1%3A2022cv01461/575368/54/
Accessed on: July 14, 2026.

UNITED STATES. Utah Court of Appeals. Garner v. Kadince, 2025 UT App 88. Utah, May 22, 2025. Available at:
https://law.justia.com/cases/utah/court-of-appeals-published/2025/20250188-ca.html
Accessed on: July 14, 2026.

UNITED STATES. Supreme Court of the State of New York. Cassata v. Michael Macrina Architect, P.C., 2026 NY Slip Op 26014. New York, January 27, 2026. Available at:
https://law.justia.com/cases/new-york/other-courts/2026/2026-ny-slip-op-26014.html
Accessed on: July 14, 2026.

UNITED STATES. Supreme Court of the State of New York. Torres v. Spraker, 2026 NY Slip Op 26077. New York, 2026. Available at:
https://law.justia.com/cases/new-york/other-courts/2026/2026-ny-slip-op-26077.html
Accessed on: July 14, 2026.

UNITED STATES. United States Court of Appeals for the Third Circuit. McCarthy v. Drug Enforcement Administration, No. 24-2704. Philadelphia, March 27, 2026. Available at:
https://law.justia.com/cases/federal/appellate-courts/ca3/24-2704/24-2704-2026-03-27.html
Accessed on: July 14, 2026.

UNITED STATES. United States Court of Appeals for the Fifth Circuit. Fletcher v. Experian Information Solutions, No. 25-20086. New Orleans, February 18, 2026. Available at:
https://law.justia.com/cases/federal/appellate-courts/ca5/25-20086/25-20086-2026-02-18.html
Accessed on: July 14, 2026.

UNITED STATES. Department of Justice. Executive Office for Immigration Review. Guidance Concerning Hallucinated or Erroneous AI-Generated Content in Filings. Washington, D.C., August 8, 2025. Available at:
https://www.justice.gov/eoir/media/1410621/dl?inline=
Accessed on: July 14, 2026.

UNITED STATES. Department of Justice. Executive Office for Immigration Review. United States v. Gold Mountain, Inc., doing business as Casa de Oro No. 4, 22 OCAHO No. 1700. Washington, D.C., April 30, 2026. Available at:
https://www.justice.gov/eoir/media/1439181/dl?inline=
Accessed on: July 14, 2026.

31.5. Canada and comparative law

CANADIAN LEGAL INFORMATION INSTITUTE. Do the Canadian Practical and Legal Frameworks on AI Meet the Needs of the Justice System? CanLII, 2025. Available at:
https://www.canlii.org/en/commentary/doc/2025CanLIIDocs1958
Accessed on: July 14, 2026.

CANADIAN LEGAL INFORMATION INSTITUTE. Artificial Intelligence & Criminal Justice: Cases and Commentary. CanLII, 2024. Available at:
https://www.canlii.org/en/commentary/doc/2024CanLIIDocs3035
Accessed on: July 14, 2026.

CANADIAN LEGAL INFORMATION INSTITUTE. Artificial Intelligence & Criminal Justice: A Primer. CanLII, 2024. Available at:
https://www.canlii.org/en/commentary/doc/2024CanLIIDocs2727
Accessed on: July 14, 2026.

CANLII BLOG. Mazaheri — Costs — Negligent and Irresponsible Use of Artificial Intelligence. Ottawa: CanLII. Available at:
https://blog.canlii.org/author/canliiadmin/
Accessed on: July 14, 2026.